net/switchdev/switchdev.c

   1 /*
   2  * net/switchdev/switchdev.c - Switch device API
   3  * Copyright (c) 2014-2015 Jiri Pirko <jiri@resnulli.us>
   4  * Copyright (c) 2014-2015 Scott Feldman <sfeldma@gmail.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/types.h>
  14 #include <linux/init.h>
  15 #include <linux/mutex.h>
  16 #include <linux/notifier.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/if_bridge.h>
  20 #include <linux/list.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/if_vlan.h>
  23 #include <linux/rtnetlink.h>
  24 #include <net/switchdev.h>
  25
  26 /**
  27  *      switchdev_trans_item_enqueue - Enqueue data item to transaction queue
  28  *
  29  *      @trans: transaction
  30  *      @data: pointer to data being queued
  31  *      @destructor: data destructor
  32  *      @tritem: transaction item being queued
  33  *
  34  *      Enqeueue data item to transaction queue. tritem is typically placed in
  35  *      cointainter pointed at by data pointer. Destructor is called on
  36  *      transaction abort and after successful commit phase in case
  37  *      the caller did not dequeue the item before.
  38  */
  39 void switchdev_trans_item_enqueue(struct switchdev_trans *trans,
  40                                   void *data, void (*destructor)(void const *),
  41                                   struct switchdev_trans_item *tritem)
  42 {
  43         tritem->data = data;
  44         tritem->destructor = destructor;
  45         list_add_tail(&tritem->list, &trans->item_list);
  46 }
  47 EXPORT_SYMBOL_GPL(switchdev_trans_item_enqueue);
  48
  49 static struct switchdev_trans_item *
  50 __switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  51 {
  52         struct switchdev_trans_item *tritem;
  53
  54         if (list_empty(&trans->item_list))
  55                 return NULL;
  56         tritem = list_first_entry(&trans->item_list,
  57                                   struct switchdev_trans_item, list);
  58         list_del(&tritem->list);
  59         return tritem;
  60 }
  61
  62 /**
  63  *      switchdev_trans_item_dequeue - Dequeue data item from transaction queue
  64  *
  65  *      @trans: transaction
  66  */
  67 void *switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  68 {
  69         struct switchdev_trans_item *tritem;
  70
  71         tritem = __switchdev_trans_item_dequeue(trans);
  72         BUG_ON(!tritem);
  73         return tritem->data;
  74 }
  75 EXPORT_SYMBOL_GPL(switchdev_trans_item_dequeue);
  76
  77 static void switchdev_trans_init(struct switchdev_trans *trans)
  78 {
  79         INIT_LIST_HEAD(&trans->item_list);
  80 }
  81
  82 static void switchdev_trans_items_destroy(struct switchdev_trans *trans)
  83 {
  84         struct switchdev_trans_item *tritem;
  85
  86         while ((tritem = __switchdev_trans_item_dequeue(trans)))
  87                 tritem->destructor(tritem->data);
  88 }
  89
  90 static void switchdev_trans_items_warn_destroy(struct net_device *dev,
  91                                                struct switchdev_trans *trans)
  92 {
  93         WARN(!list_empty(&trans->item_list), "%s: transaction item queue is not empty.\n",
  94              dev->name);
  95         switchdev_trans_items_destroy(trans);
  96 }
  97
  98 static LIST_HEAD(deferred);
  99 static DEFINE_SPINLOCK(deferred_lock);
 100
 101 typedef void switchdev_deferred_func_t(struct net_device *dev,
 102                                        const void *data);
 103
 104 struct switchdev_deferred_item {
 105         struct list_head list;
 106         struct net_device *dev;
 107         switchdev_deferred_func_t *func;
 108         unsigned long data[0];
 109 };
 110
 111 static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
 112 {
 113         struct switchdev_deferred_item *dfitem;
 114
 115         spin_lock_bh(&deferred_lock);
 116         if (list_empty(&deferred)) {
 117                 dfitem = NULL;
 118                 goto unlock;
 119         }
 120         dfitem = list_first_entry(&deferred,
 121                                   struct switchdev_deferred_item, list);
 122         list_del(&dfitem->list);
 123 unlock:
 124         spin_unlock_bh(&deferred_lock);
 125         return dfitem;
 126 }
 127
 128 /**
 129  *      switchdev_deferred_process - Process ops in deferred queue
 130  *
 131  *      Called to flush the ops currently queued in deferred ops queue.
 132  *      rtnl_lock must be held.
 133  */
 134 void switchdev_deferred_process(void)
 135 {
 136         struct switchdev_deferred_item *dfitem;
 137
 138         ASSERT_RTNL();
 139
 140         while ((dfitem = switchdev_deferred_dequeue())) {
 141                 dfitem->func(dfitem->dev, dfitem->data);
 142                 dev_put(dfitem->dev);
 143                 kfree(dfitem);
 144         }
 145 }
 146 EXPORT_SYMBOL_GPL(switchdev_deferred_process);
 147
 148 static void switchdev_deferred_process_work(struct work_struct *work)
 149 {
 150         rtnl_lock();
 151         switchdev_deferred_process();
 152         rtnl_unlock();
 153 }
 154
 155 static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
 156
 157 static int switchdev_deferred_enqueue(struct net_device *dev,
 158                                       const void *data, size_t data_len,
 159                                       switchdev_deferred_func_t *func)
 160 {
 161         struct switchdev_deferred_item *dfitem;
 162
 163         dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
 164         if (!dfitem)
 165                 return -ENOMEM;
 166         dfitem->dev = dev;
 167         dfitem->func = func;
 168         memcpy(dfitem->data, data, data_len);
 169         dev_hold(dev);
 170         spin_lock_bh(&deferred_lock);
 171         list_add_tail(&dfitem->list, &deferred);
 172         spin_unlock_bh(&deferred_lock);
 173         schedule_work(&deferred_process_work);
 174         return 0;
 175 }
 176
 177 /**
 178  *      switchdev_port_attr_get - Get port attribute
 179  *
 180  *      @dev: port device
 181  *      @attr: attribute to get
 182  */
 183 int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
 184 {
 185         const struct switchdev_ops *ops = dev->switchdev_ops;
 186         struct net_device *lower_dev;
 187         struct list_head *iter;
 188         struct switchdev_attr first = {
 189                 .id = SWITCHDEV_ATTR_ID_UNDEFINED
 190         };
 191         int err = -EOPNOTSUPP;
 192
 193         if (ops && ops->switchdev_port_attr_get)
 194                 return ops->switchdev_port_attr_get(dev, attr);
 195
 196         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 197                 return err;
 198
 199         /* Switch device port(s) may be stacked under
 200          * bond/team/vlan dev, so recurse down to get attr on
 201          * each port.  Return -ENODATA if attr values don't
 202          * compare across ports.
 203          */
 204
 205         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 206                 err = switchdev_port_attr_get(lower_dev, attr);
 207                 if (err)
 208                         break;
 209                 if (first.id == SWITCHDEV_ATTR_ID_UNDEFINED)
 210                         first = *attr;
 211                 else if (memcmp(&first, attr, sizeof(*attr)))
 212                         return -ENODATA;
 213         }
 214
 215         return err;
 216 }
 217 EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
 218
 219 static int __switchdev_port_attr_set(struct net_device *dev,
 220                                      const struct switchdev_attr *attr,
 221                                      struct switchdev_trans *trans)
 222 {
 223         const struct switchdev_ops *ops = dev->switchdev_ops;
 224         struct net_device *lower_dev;
 225         struct list_head *iter;
 226         int err = -EOPNOTSUPP;
 227
 228         if (ops && ops->switchdev_port_attr_set) {
 229                 err = ops->switchdev_port_attr_set(dev, attr, trans);
 230                 goto done;
 231         }
 232
 233         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 234                 goto done;
 235
 236         /* Switch device port(s) may be stacked under
 237          * bond/team/vlan dev, so recurse down to set attr on
 238          * each port.
 239          */
 240
 241         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 242                 err = __switchdev_port_attr_set(lower_dev, attr, trans);
 243                 if (err)
 244                         break;
 245         }
 246
 247 done:
 248         if (err == -EOPNOTSUPP && attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
 249                 err = 0;
 250
 251         return err;
 252 }
 253
 254 static int switchdev_port_attr_set_now(struct net_device *dev,
 255                                        const struct switchdev_attr *attr)
 256 {
 257         struct switchdev_trans trans;
 258         int err;
 259
 260         switchdev_trans_init(&trans);
 261
 262         /* Phase I: prepare for attr set. Driver/device should fail
 263          * here if there are going to be issues in the commit phase,
 264          * such as lack of resources or support.  The driver/device
 265          * should reserve resources needed for the commit phase here,
 266          * but should not commit the attr.
 267          */
 268
 269         trans.ph_prepare = true;
 270         err = __switchdev_port_attr_set(dev, attr, &trans);
 271         if (err) {
 272                 /* Prepare phase failed: abort the transaction.  Any
 273                  * resources reserved in the prepare phase are
 274                  * released.
 275                  */
 276
 277                 if (err != -EOPNOTSUPP)
 278                         switchdev_trans_items_destroy(&trans);
 279
 280                 return err;
 281         }
 282
 283         /* Phase II: commit attr set.  This cannot fail as a fault
 284          * of driver/device.  If it does, it's a bug in the driver/device
 285          * because the driver said everythings was OK in phase I.
 286          */
 287
 288         trans.ph_prepare = false;
 289         err = __switchdev_port_attr_set(dev, attr, &trans);
 290         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 291              dev->name, attr->id);
 292         switchdev_trans_items_warn_destroy(dev, &trans);
 293
 294         return err;
 295 }
 296
 297 static void switchdev_port_attr_set_deferred(struct net_device *dev,
 298                                              const void *data)
 299 {
 300         const struct switchdev_attr *attr = data;
 301         int err;
 302
 303         err = switchdev_port_attr_set_now(dev, attr);
 304         if (err && err != -EOPNOTSUPP)
 305                 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
 306                            err, attr->id);
 307         if (attr->complete)
 308                 attr->complete(dev, err, attr->complete_priv);
 309 }
 310
 311 static int switchdev_port_attr_set_defer(struct net_device *dev,
 312                                          const struct switchdev_attr *attr)
 313 {
 314         return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
 315                                           switchdev_port_attr_set_deferred);
 316 }
 317
 318 /**
 319  *      switchdev_port_attr_set - Set port attribute
 320  *
 321  *      @dev: port device
 322  *      @attr: attribute to set
 323  *
 324  *      Use a 2-phase prepare-commit transaction model to ensure
 325  *      system is not left in a partially updated state due to
 326  *      failure from driver/device.
 327  *
 328  *      rtnl_lock must be held and must not be in atomic section,
 329  *      in case SWITCHDEV_F_DEFER flag is not set.
 330  */
 331 int switchdev_port_attr_set(struct net_device *dev,
 332                             const struct switchdev_attr *attr)
 333 {
 334         if (attr->flags & SWITCHDEV_F_DEFER)
 335                 return switchdev_port_attr_set_defer(dev, attr);
 336         ASSERT_RTNL();
 337         return switchdev_port_attr_set_now(dev, attr);
 338 }
 339 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 340
 341 static size_t switchdev_obj_size(const struct switchdev_obj *obj)
 342 {
 343         switch (obj->id) {
 344         case SWITCHDEV_OBJ_ID_PORT_VLAN:
 345                 return sizeof(struct switchdev_obj_port_vlan);
 346         case SWITCHDEV_OBJ_ID_PORT_FDB:
 347                 return sizeof(struct switchdev_obj_port_fdb);
 348         case SWITCHDEV_OBJ_ID_PORT_MDB:
 349                 return sizeof(struct switchdev_obj_port_mdb);
 350         default:
 351                 BUG();
 352         }
 353         return 0;
 354 }
 355
 356 static int __switchdev_port_obj_add(struct net_device *dev,
 357                                     const struct switchdev_obj *obj,
 358                                     struct switchdev_trans *trans)
 359 {
 360         const struct switchdev_ops *ops = dev->switchdev_ops;
 361         struct net_device *lower_dev;
 362         struct list_head *iter;
 363         int err = -EOPNOTSUPP;
 364
 365         if (ops && ops->switchdev_port_obj_add)
 366                 return ops->switchdev_port_obj_add(dev, obj, trans);
 367
 368         /* Switch device port(s) may be stacked under
 369          * bond/team/vlan dev, so recurse down to add object on
 370          * each port.
 371          */
 372
 373         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 374                 err = __switchdev_port_obj_add(lower_dev, obj, trans);
 375                 if (err)
 376                         break;
 377         }
 378
 379         return err;
 380 }
 381
 382 static int switchdev_port_obj_add_now(struct net_device *dev,
 383                                       const struct switchdev_obj *obj)
 384 {
 385         struct switchdev_trans trans;
 386         int err;
 387
 388         ASSERT_RTNL();
 389
 390         switchdev_trans_init(&trans);
 391
 392         /* Phase I: prepare for obj add. Driver/device should fail
 393          * here if there are going to be issues in the commit phase,
 394          * such as lack of resources or support.  The driver/device
 395          * should reserve resources needed for the commit phase here,
 396          * but should not commit the obj.
 397          */
 398
 399         trans.ph_prepare = true;
 400         err = __switchdev_port_obj_add(dev, obj, &trans);
 401         if (err) {
 402                 /* Prepare phase failed: abort the transaction.  Any
 403                  * resources reserved in the prepare phase are
 404                  * released.
 405                  */
 406
 407                 if (err != -EOPNOTSUPP)
 408                         switchdev_trans_items_destroy(&trans);
 409
 410                 return err;
 411         }
 412
 413         /* Phase II: commit obj add.  This cannot fail as a fault
 414          * of driver/device.  If it does, it's a bug in the driver/device
 415          * because the driver said everythings was OK in phase I.
 416          */
 417
 418         trans.ph_prepare = false;
 419         err = __switchdev_port_obj_add(dev, obj, &trans);
 420         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 421         switchdev_trans_items_warn_destroy(dev, &trans);
 422
 423         return err;
 424 }
 425
 426 static void switchdev_port_obj_add_deferred(struct net_device *dev,
 427                                             const void *data)
 428 {
 429         const struct switchdev_obj *obj = data;
 430         int err;
 431
 432         err = switchdev_port_obj_add_now(dev, obj);
 433         if (err && err != -EOPNOTSUPP)
 434                 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
 435                            err, obj->id);
 436         if (obj->complete)
 437                 obj->complete(dev, err, obj->complete_priv);
 438 }
 439
 440 static int switchdev_port_obj_add_defer(struct net_device *dev,
 441                                         const struct switchdev_obj *obj)
 442 {
 443         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 444                                           switchdev_port_obj_add_deferred);
 445 }
 446
 447 /**
 448  *      switchdev_port_obj_add - Add port object
 449  *
 450  *      @dev: port device
 451  *      @id: object ID
 452  *      @obj: object to add
 453  *
 454  *      Use a 2-phase prepare-commit transaction model to ensure
 455  *      system is not left in a partially updated state due to
 456  *      failure from driver/device.
 457  *
 458  *      rtnl_lock must be held and must not be in atomic section,
 459  *      in case SWITCHDEV_F_DEFER flag is not set.
 460  */
 461 int switchdev_port_obj_add(struct net_device *dev,
 462                            const struct switchdev_obj *obj)
 463 {
 464         if (obj->flags & SWITCHDEV_F_DEFER)
 465                 return switchdev_port_obj_add_defer(dev, obj);
 466         ASSERT_RTNL();
 467         return switchdev_port_obj_add_now(dev, obj);
 468 }
 469 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 470
 471 static int switchdev_port_obj_del_now(struct net_device *dev,
 472                                       const struct switchdev_obj *obj)
 473 {
 474         const struct switchdev_ops *ops = dev->switchdev_ops;
 475         struct net_device *lower_dev;
 476         struct list_head *iter;
 477         int err = -EOPNOTSUPP;
 478
 479         if (ops && ops->switchdev_port_obj_del)
 480                 return ops->switchdev_port_obj_del(dev, obj);
 481
 482         /* Switch device port(s) may be stacked under
 483          * bond/team/vlan dev, so recurse down to delete object on
 484          * each port.
 485          */
 486
 487         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 488                 err = switchdev_port_obj_del_now(lower_dev, obj);
 489                 if (err)
 490                         break;
 491         }
 492
 493         return err;
 494 }
 495
 496 static void switchdev_port_obj_del_deferred(struct net_device *dev,
 497                                             const void *data)
 498 {
 499         const struct switchdev_obj *obj = data;
 500         int err;
 501
 502         err = switchdev_port_obj_del_now(dev, obj);
 503         if (err && err != -EOPNOTSUPP)
 504                 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
 505                            err, obj->id);
 506         if (obj->complete)
 507                 obj->complete(dev, err, obj->complete_priv);
 508 }
 509
 510 static int switchdev_port_obj_del_defer(struct net_device *dev,
 511                                         const struct switchdev_obj *obj)
 512 {
 513         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 514                                           switchdev_port_obj_del_deferred);
 515 }
 516
 517 /**
 518  *      switchdev_port_obj_del - Delete port object
 519  *
 520  *      @dev: port device
 521  *      @id: object ID
 522  *      @obj: object to delete
 523  *
 524  *      rtnl_lock must be held and must not be in atomic section,
 525  *      in case SWITCHDEV_F_DEFER flag is not set.
 526  */
 527 int switchdev_port_obj_del(struct net_device *dev,
 528                            const struct switchdev_obj *obj)
 529 {
 530         if (obj->flags & SWITCHDEV_F_DEFER)
 531                 return switchdev_port_obj_del_defer(dev, obj);
 532         ASSERT_RTNL();
 533         return switchdev_port_obj_del_now(dev, obj);
 534 }
 535 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 536
 537 /**
 538  *      switchdev_port_obj_dump - Dump port objects
 539  *
 540  *      @dev: port device
 541  *      @id: object ID
 542  *      @obj: object to dump
 543  *      @cb: function to call with a filled object
 544  *
 545  *      rtnl_lock must be held.
 546  */
 547 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj,
 548                             switchdev_obj_dump_cb_t *cb)
 549 {
 550         const struct switchdev_ops *ops = dev->switchdev_ops;
 551         struct net_device *lower_dev;
 552         struct list_head *iter;
 553         int err = -EOPNOTSUPP;
 554
 555         ASSERT_RTNL();
 556
 557         if (ops && ops->switchdev_port_obj_dump)
 558                 return ops->switchdev_port_obj_dump(dev, obj, cb);
 559
 560         /* Switch device port(s) may be stacked under
 561          * bond/team/vlan dev, so recurse down to dump objects on
 562          * first port at bottom of stack.
 563          */
 564
 565         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 566                 err = switchdev_port_obj_dump(lower_dev, obj, cb);
 567                 break;
 568         }
 569
 570         return err;
 571 }
 572 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
 573
 574 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
 575
 576 /**
 577  *      register_switchdev_notifier - Register notifier
 578  *      @nb: notifier_block
 579  *
 580  *      Register switch device notifier.
 581  */
 582 int register_switchdev_notifier(struct notifier_block *nb)
 583 {
 584         return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
 585 }
 586 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 587
 588 /**
 589  *      unregister_switchdev_notifier - Unregister notifier
 590  *      @nb: notifier_block
 591  *
 592  *      Unregister switch device notifier.
 593  */
 594 int unregister_switchdev_notifier(struct notifier_block *nb)
 595 {
 596         return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
 597 }
 598 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 599
 600 /**
 601  *      call_switchdev_notifiers - Call notifiers
 602  *      @val: value passed unmodified to notifier function
 603  *      @dev: port device
 604  *      @info: notifier information data
 605  *
 606  *      Call all network notifier blocks.
 607  */
 608 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 609                              struct switchdev_notifier_info *info)
 610 {
 611         info->dev = dev;
 612         return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
 613 }
 614 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 615
 616 struct switchdev_vlan_dump {
 617         struct switchdev_obj_port_vlan vlan;
 618         struct sk_buff *skb;
 619         u32 filter_mask;
 620         u16 flags;
 621         u16 begin;
 622         u16 end;
 623 };
 624
 625 static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
 626 {
 627         struct bridge_vlan_info vinfo;
 628
 629         vinfo.flags = dump->flags;
 630
 631         if (dump->begin == 0 && dump->end == 0) {
 632                 return 0;
 633         } else if (dump->begin == dump->end) {
 634                 vinfo.vid = dump->begin;
 635                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 636                             sizeof(vinfo), &vinfo))
 637                         return -EMSGSIZE;
 638         } else {
 639                 vinfo.vid = dump->begin;
 640                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
 641                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 642                             sizeof(vinfo), &vinfo))
 643                         return -EMSGSIZE;
 644                 vinfo.vid = dump->end;
 645                 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
 646                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
 647                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 648                             sizeof(vinfo), &vinfo))
 649                         return -EMSGSIZE;
 650         }
 651
 652         return 0;
 653 }
 654
 655 static int switchdev_port_vlan_dump_cb(struct switchdev_obj *obj)
 656 {
 657         struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
 658         struct switchdev_vlan_dump *dump =
 659                 container_of(vlan, struct switchdev_vlan_dump, vlan);
 660         int err = 0;
 661
 662         if (vlan->vid_begin > vlan->vid_end)
 663                 return -EINVAL;
 664
 665         if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
 666                 dump->flags = vlan->flags;
 667                 for (dump->begin = dump->end = vlan->vid_begin;
 668                      dump->begin <= vlan->vid_end;
 669                      dump->begin++, dump->end++) {
 670                         err = switchdev_port_vlan_dump_put(dump);
 671                         if (err)
 672                                 return err;
 673                 }
 674         } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
 675                 if (dump->begin > vlan->vid_begin &&
 676                     dump->begin >= vlan->vid_end) {
 677                         if ((dump->begin - 1) == vlan->vid_end &&
 678                             dump->flags == vlan->flags) {
 679                                 /* prepend */
 680                                 dump->begin = vlan->vid_begin;
 681                         } else {
 682                                 err = switchdev_port_vlan_dump_put(dump);
 683                                 dump->flags = vlan->flags;
 684                                 dump->begin = vlan->vid_begin;
 685                                 dump->end = vlan->vid_end;
 686                         }
 687                 } else if (dump->end <= vlan->vid_begin &&
 688                            dump->end < vlan->vid_end) {
 689                         if ((dump->end  + 1) == vlan->vid_begin &&
 690                             dump->flags == vlan->flags) {
 691                                 /* append */
 692                                 dump->end = vlan->vid_end;
 693                         } else {
 694                                 err = switchdev_port_vlan_dump_put(dump);
 695                                 dump->flags = vlan->flags;
 696                                 dump->begin = vlan->vid_begin;
 697                                 dump->end = vlan->vid_end;
 698                         }
 699                 } else {
 700                         err = -EINVAL;
 701                 }
 702         }
 703
 704         return err;
 705 }
 706
 707 static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
 708                                     u32 filter_mask)
 709 {
 710         struct switchdev_vlan_dump dump = {
 711                 .vlan.obj.orig_dev = dev,
 712                 .vlan.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 713                 .skb = skb,
 714                 .filter_mask = filter_mask,
 715         };
 716         int err = 0;
 717
 718         if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
 719             (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
 720                 err = switchdev_port_obj_dump(dev, &dump.vlan.obj,
 721                                               switchdev_port_vlan_dump_cb);
 722                 if (err)
 723                         goto err_out;
 724                 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
 725                         /* last one */
 726                         err = switchdev_port_vlan_dump_put(&dump);
 727         }
 728
 729 err_out:
 730         return err == -EOPNOTSUPP ? 0 : err;
 731 }
 732
 733 /**
 734  *      switchdev_port_bridge_getlink - Get bridge port attributes
 735  *
 736  *      @dev: port device
 737  *
 738  *      Called for SELF on rtnl_bridge_getlink to get bridge port
 739  *      attributes.
 740  */
 741 int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
 742                                   struct net_device *dev, u32 filter_mask,
 743                                   int nlflags)
 744 {
 745         struct switchdev_attr attr = {
 746                 .orig_dev = dev,
 747                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 748         };
 749         u16 mode = BRIDGE_MODE_UNDEF;
 750         u32 mask = BR_LEARNING | BR_LEARNING_SYNC | BR_FLOOD;
 751         int err;
 752
 753         if (!netif_is_bridge_port(dev))
 754                 return -EOPNOTSUPP;
 755
 756         err = switchdev_port_attr_get(dev, &attr);
 757         if (err && err != -EOPNOTSUPP)
 758                 return err;
 759
 760         return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
 761                                        attr.u.brport_flags, mask, nlflags,
 762                                        filter_mask, switchdev_port_vlan_fill);
 763 }
 764 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
 765
 766 static int switchdev_port_br_setflag(struct net_device *dev,
 767                                      struct nlattr *nlattr,
 768                                      unsigned long brport_flag)
 769 {
 770         struct switchdev_attr attr = {
 771                 .orig_dev = dev,
 772                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 773         };
 774         u8 flag = nla_get_u8(nlattr);
 775         int err;
 776
 777         err = switchdev_port_attr_get(dev, &attr);
 778         if (err)
 779                 return err;
 780
 781         if (flag)
 782                 attr.u.brport_flags |= brport_flag;
 783         else
 784                 attr.u.brport_flags &= ~brport_flag;
 785
 786         return switchdev_port_attr_set(dev, &attr);
 787 }
 788
 789 static const struct nla_policy
 790 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
 791         [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
 792         [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
 793         [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
 794         [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
 795         [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
 796         [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
 797         [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
 798         [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
 799         [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
 800         [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
 801 };
 802
 803 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
 804                                               struct nlattr *protinfo)
 805 {
 806         struct nlattr *attr;
 807         int rem;
 808         int err;
 809
 810         err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
 811                                   switchdev_port_bridge_policy, NULL);
 812         if (err)
 813                 return err;
 814
 815         nla_for_each_nested(attr, protinfo, rem) {
 816                 switch (nla_type(attr)) {
 817                 case IFLA_BRPORT_LEARNING:
 818                         err = switchdev_port_br_setflag(dev, attr,
 819                                                         BR_LEARNING);
 820                         break;
 821                 case IFLA_BRPORT_LEARNING_SYNC:
 822                         err = switchdev_port_br_setflag(dev, attr,
 823                                                         BR_LEARNING_SYNC);
 824                         break;
 825                 case IFLA_BRPORT_UNICAST_FLOOD:
 826                         err = switchdev_port_br_setflag(dev, attr, BR_FLOOD);
 827                         break;
 828                 default:
 829                         err = -EOPNOTSUPP;
 830                         break;
 831                 }
 832                 if (err)
 833                         return err;
 834         }
 835
 836         return 0;
 837 }
 838
 839 static int switchdev_port_br_afspec(struct net_device *dev,
 840                                     struct nlattr *afspec,
 841                                     int (*f)(struct net_device *dev,
 842                                              const struct switchdev_obj *obj))
 843 {
 844         struct nlattr *attr;
 845         struct bridge_vlan_info *vinfo;
 846         struct switchdev_obj_port_vlan vlan = {
 847                 .obj.orig_dev = dev,
 848                 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 849         };
 850         int rem;
 851         int err;
 852
 853         nla_for_each_nested(attr, afspec, rem) {
 854                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 855                         continue;
 856                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 857                         return -EINVAL;
 858                 vinfo = nla_data(attr);
 859                 if (!vinfo->vid || vinfo->vid >= VLAN_VID_MASK)
 860                         return -EINVAL;
 861                 vlan.flags = vinfo->flags;
 862                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 863                         if (vlan.vid_begin)
 864                                 return -EINVAL;
 865                         vlan.vid_begin = vinfo->vid;
 866                         /* don't allow range of pvids */
 867                         if (vlan.flags & BRIDGE_VLAN_INFO_PVID)
 868                                 return -EINVAL;
 869                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 870                         if (!vlan.vid_begin)
 871                                 return -EINVAL;
 872                         vlan.vid_end = vinfo->vid;
 873                         if (vlan.vid_end <= vlan.vid_begin)
 874                                 return -EINVAL;
 875                         err = f(dev, &vlan.obj);
 876                         if (err)
 877                                 return err;
 878                         vlan.vid_begin = 0;
 879                 } else {
 880                         if (vlan.vid_begin)
 881                                 return -EINVAL;
 882                         vlan.vid_begin = vinfo->vid;
 883                         vlan.vid_end = vinfo->vid;
 884                         err = f(dev, &vlan.obj);
 885                         if (err)
 886                                 return err;
 887                         vlan.vid_begin = 0;
 888                 }
 889         }
 890
 891         return 0;
 892 }
 893
 894 /**
 895  *      switchdev_port_bridge_setlink - Set bridge port attributes
 896  *
 897  *      @dev: port device
 898  *      @nlh: netlink header
 899  *      @flags: netlink flags
 900  *
 901  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 902  *      attributes.
 903  */
 904 int switchdev_port_bridge_setlink(struct net_device *dev,
 905                                   struct nlmsghdr *nlh, u16 flags)
 906 {
 907         struct nlattr *protinfo;
 908         struct nlattr *afspec;
 909         int err = 0;
 910
 911         if (!netif_is_bridge_port(dev))
 912                 return -EOPNOTSUPP;
 913
 914         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 915                                    IFLA_PROTINFO);
 916         if (protinfo) {
 917                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 918                 if (err)
 919                         return err;
 920         }
 921
 922         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 923                                  IFLA_AF_SPEC);
 924         if (afspec)
 925                 err = switchdev_port_br_afspec(dev, afspec,
 926                                                switchdev_port_obj_add);
 927
 928         return err;
 929 }
 930 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 931
 932 /**
 933  *      switchdev_port_bridge_dellink - Set bridge port attributes
 934  *
 935  *      @dev: port device
 936  *      @nlh: netlink header
 937  *      @flags: netlink flags
 938  *
 939  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 940  *      attributes.
 941  */
 942 int switchdev_port_bridge_dellink(struct net_device *dev,
 943                                   struct nlmsghdr *nlh, u16 flags)
 944 {
 945         struct nlattr *afspec;
 946
 947         if (!netif_is_bridge_port(dev))
 948                 return -EOPNOTSUPP;
 949
 950         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 951                                  IFLA_AF_SPEC);
 952         if (afspec)
 953                 return switchdev_port_br_afspec(dev, afspec,
 954                                                 switchdev_port_obj_del);
 955
 956         return 0;
 957 }
 958 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 959
 960 /**
 961  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 962  *
 963  *      @ndmsg: netlink hdr
 964  *      @nlattr: netlink attributes
 965  *      @dev: port device
 966  *      @addr: MAC address to add
 967  *      @vid: VLAN to add
 968  *
 969  *      Add FDB entry to switch device.
 970  */
 971 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 972                            struct net_device *dev, const unsigned char *addr,
 973                            u16 vid, u16 nlm_flags)
 974 {
 975         struct switchdev_obj_port_fdb fdb = {
 976                 .obj.orig_dev = dev,
 977                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 978                 .vid = vid,
 979         };
 980
 981         ether_addr_copy(fdb.addr, addr);
 982         return switchdev_port_obj_add(dev, &fdb.obj);
 983 }
 984 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 985
 986 /**
 987  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 988  *
 989  *      @ndmsg: netlink hdr
 990  *      @nlattr: netlink attributes
 991  *      @dev: port device
 992  *      @addr: MAC address to delete
 993  *      @vid: VLAN to delete
 994  *
 995  *      Delete FDB entry from switch device.
 996  */
 997 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 998                            struct net_device *dev, const unsigned char *addr,
 999                            u16 vid)
1000 {
1001         struct switchdev_obj_port_fdb fdb = {
1002                 .obj.orig_dev = dev,
1003                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1004                 .vid = vid,
1005         };
1006
1007         ether_addr_copy(fdb.addr, addr);
1008         return switchdev_port_obj_del(dev, &fdb.obj);
1009 }
1010 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
1011
1012 struct switchdev_fdb_dump {
1013         struct switchdev_obj_port_fdb fdb;
1014         struct net_device *dev;
1015         struct sk_buff *skb;
1016         struct netlink_callback *cb;
1017         int idx;
1018 };
1019
1020 static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
1021 {
1022         struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
1023         struct switchdev_fdb_dump *dump =
1024                 container_of(fdb, struct switchdev_fdb_dump, fdb);
1025         u32 portid = NETLINK_CB(dump->cb->skb).portid;
1026         u32 seq = dump->cb->nlh->nlmsg_seq;
1027         struct nlmsghdr *nlh;
1028         struct ndmsg *ndm;
1029
1030         if (dump->idx < dump->cb->args[2])
1031                 goto skip;
1032
1033         nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
1034                         sizeof(*ndm), NLM_F_MULTI);
1035         if (!nlh)
1036                 return -EMSGSIZE;
1037
1038         ndm = nlmsg_data(nlh);
1039         ndm->ndm_family  = AF_BRIDGE;
1040         ndm->ndm_pad1    = 0;
1041         ndm->ndm_pad2    = 0;
1042         ndm->ndm_flags   = NTF_SELF;
1043         ndm->ndm_type    = 0;
1044         ndm->ndm_ifindex = dump->dev->ifindex;
1045         ndm->ndm_state   = fdb->ndm_state;
1046
1047         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
1048                 goto nla_put_failure;
1049
1050         if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, fdb->vid))
1051                 goto nla_put_failure;
1052
1053         nlmsg_end(dump->skb, nlh);
1054
1055 skip:
1056         dump->idx++;
1057         return 0;
1058
1059 nla_put_failure:
1060         nlmsg_cancel(dump->skb, nlh);
1061         return -EMSGSIZE;
1062 }
1063
1064 /**
1065  *      switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
1066  *
1067  *      @skb: netlink skb
1068  *      @cb: netlink callback
1069  *      @dev: port device
1070  *      @filter_dev: filter device
1071  *      @idx:
1072  *
1073  *      Dump FDB entries from switch device.
1074  */
1075 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1076                             struct net_device *dev,
1077                             struct net_device *filter_dev, int *idx)
1078 {
1079         struct switchdev_fdb_dump dump = {
1080                 .fdb.obj.orig_dev = dev,
1081                 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1082                 .dev = dev,
1083                 .skb = skb,
1084                 .cb = cb,
1085                 .idx = *idx,
1086         };
1087         int err;
1088
1089         err = switchdev_port_obj_dump(dev, &dump.fdb.obj,
1090                                       switchdev_port_fdb_dump_cb);
1091         *idx = dump.idx;
1092         return err;
1093 }
1094 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
1095
1096 bool switchdev_port_same_parent_id(struct net_device *a,
1097                                    struct net_device *b)
1098 {
1099         struct switchdev_attr a_attr = {
1100                 .orig_dev = a,
1101                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1102         };
1103         struct switchdev_attr b_attr = {
1104                 .orig_dev = b,
1105                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1106         };
1107
1108         if (switchdev_port_attr_get(a, &a_attr) ||
1109             switchdev_port_attr_get(b, &b_attr))
1110                 return false;
1111
1112         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1113 }
1114 EXPORT_SYMBOL_GPL(switchdev_port_same_parent_id);